[
https://issues.apache.org/jira/browse/TS-4846?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15480806#comment-15480806
]
Leif Hedstrom commented on TS-4846:
-----------------------------------
I'm not opposed to this, but also not sure I see the real value of requiring
this until we have some hard dependencies on the library itself. This also
brings us into the pain of dealing with jemalloc versions if we want to code to
some of their APIs, lowest common denominator seems to be jemalloc v3.6.1, with
v4.1.x being available on modern distros.
Two more thoughts:
1) If we do this, we ought to remove the support for tcmalloc.
2) Should we consider importing jemalloc into our lib/ tree?
I'm -0 on this for now, at least until we have a real need for this dependency.
However, in preparation, we could remove the (explicit) support for tcmalloc
for v7.0.0, that'll at least make it easier to add new features with jemalloc
specific requirements.
> Make jemalloc a required dependency.
> ------------------------------------
>
> Key: TS-4846
> URL: https://issues.apache.org/jira/browse/TS-4846
> Project: Traffic Server
> Issue Type: New Feature
> Components: Core, Performance
> Reporter: James Peach
>
> If we make {{jemalloc}} a required dependency, then all operators would
> better better performance without any special configuration. We can also then
> build tooling that used the jemalloc [heap
> profiling|https://github.com/jemalloc/jemalloc/wiki/Use-Case%3A-Heap-Profiling]
> and
> [statistics|https://github.com/jemalloc/jemalloc/wiki/Use-Case%3A-Basic-Allocator-Statistics]
> support to get more information about memory and allocation patterns.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
